Cloudflare's November 18 Outage, One Year Later: 524 Status Pages Named It
On November 18, 2025, 524 companies opened a status-page incident that named Cloudflare. When they posted, how long they stayed red, and what to change.
524 companies, one bad file
Cloudflare's core proxy started failing at 11:20 UTC on November 18, 2025, and Cloudflare says every system was back to normal at 17:06 UTC. In that window, 524 companies in PulsAPI's catalog opened an incident on their own status page that names Cloudflare. On a normal day that number is zero. On the rare day it isn't, it is usually one or two.
Cloudflare's postmortem explains what broke inside Cloudflare. The 524 vendor posts show what happened to everyone downstream, and the timing is the interesting part. The median company told its customers 89 minutes after the failures began. A quarter took longer than two and a half hours. 74 opened their incident after 14:30, when Cloudflare says its core traffic was already flowing normally. And at 17:00, about when Cloudflare declared everything fixed, only half of the vendors had closed theirs.
This is for teams that run on Cloudflare, and for the larger group that runs on vendors who do. Live status is on Cloudflare status, and the longer record is in the Cloudflare outage history.
| Companies whose incident named Cloudflare | 524 |
| Companies naming Cloudflare on a typical day (Oct 2025 to Jan 2026) | 0 on most days, never more than 14 |
| First vendor post | 11:30 UTC |
| Vendors that posted before Cloudflare opened its incident call (11:35) | 4 |
| Median time from first failures to the vendor's post | 89 minutes |
| Vendors that first posted after Cloudflare's core traffic recovered (14:30) | 74 |
| Vendors that had closed their incident by 17:00 | 256 of 508 |
| Vendors still open the next day or later | 45 of 508 |
| Median length of a vendor's incident | 3 h 38 min |
What broke inside Cloudflare
Cloudflare published its account the same day: Cloudflare outage on November 18, 2025, by CEO Matthew Prince. At 11:05 a permissions change on a ClickHouse database cluster made a metadata query return duplicate rows. Those rows went into a feature file that Cloudflare's Bot Management system rebuilds every few minutes, and the file roughly doubled. The proxy that loads it had a hard limit of 200 features. The new file went over it, the proxy code panicked, and sites behind Cloudflare started returning HTTP 5xx errors.
For the first stretch the failure came and went. Only some database nodes had the change, so the file was sometimes built good and sometimes bad, and the network flipped between working and broken every five minutes. Cloudflare says that pattern, plus its status page going down at the same moment, made the team suspect an attack at first. The status page outage turned out to be a coincidence. It wasn't an attack either.
| 11:05 | Database permissions change deployed |
| 11:20 | Network starts failing to deliver core traffic |
| 11:31 | First automated test detects the problem |
| 11:35 | Incident call opened |
| 13:05 | Workers KV and Access bypass reduces impact |
| 14:24 | Bad feature files stop being created and pushed |
| 14:30 | Core traffic largely flowing as normal |
| 17:06 | All Cloudflare systems functioning as normal |
How fast vendors told their customers
The earliest vendor post in our data is from Sword, at 11:30:18 UTC, ten minutes in. Woosmap followed nine seconds later. Exercise.com and HetrixTools also posted before 11:35, the minute Cloudflare opened its own incident call. Those four were early. Most were not.
An hour after the failures started, 85 of the 524 had said anything. The big wave came in the next hour: 205 companies opened an incident between 12:00 and 12:59, more than in any other hour of the day. The median post went up at 12:49. By 13:00 a little over half had posted.
Ninety minutes is roughly what it takes when the cause is somewhere else. Someone has to notice, rule out their own last deploy, work out that the problem is Cloudflare, and get a sentence past whoever approves status updates. None of that is negligence. It does mean a vendor's status page is a slow signal at the start of an upstream outage, which is exactly when you want a fast one.
| 11:00 to 11:59 | 85 |
| 12:00 to 12:59 | 205 |
| 13:00 to 13:59 | 106 |
| 14:00 to 14:59 | 92 |
| 15:00 to 15:59 | 21 |
| 16:00 to 16:59 | 6 |
| 17:00 to 23:59 | 9 |
Who named Cloudflare
Most of them were not infrastructure companies. In PulsAPI's catalog, 248 of the 524 are general SaaS products, 59 are fintech, 38 are marketing tools and 33 are developer tools. 23 are AI companies. Collaboration software, payment apps and course platforms all posted about the same failure, and most of their customers had no reason to know a CDN was involved.
| SaaS | 248 |
| Fintech | 59 |
| Marketing | 38 |
| Developer tools | 33 |
| Security | 26 |
| AI and machine learning | 23 |
| Communication | 21 |
| Cloud and hosting | 18 |
| Analytics | 16 |
| Monitoring | 12 |
| Media | 10 |
| Identity | 9 |
| Database | 6 |
| Other | 5 |
Fifteen well-known names, in their own words
The table below takes fifteen of the better-known companies and shows when each one posted, when it closed the incident, and the severity it chose. The titles are copied as published, typos included.
Look at the severity column before drawing any conclusions from it. Udemy called its incident critical, Datadog called delayed monitor notifications minor, and Figma set no severity at all. Each label is the vendor's own call about how to describe it, not a measurement of how broken the product was. Some of the 91 posts with no severity were notices that a product was fine. Amplitude's title, for example, reads "Cloudflare Outage: Amplitude User Interface and Data Ingestion Unaffected".
| Groq | 11:52 | 15:26 | Major | Console, Website, & API Availability Issues: Global Cloudflare Outage |
| StatusCake | 12:02 | 14:58 | Major | Cloudflare outage |
| Canva | 12:04 | 14:56 | Major | Users are experiencing issues loading Canva |
| Pipedrive | 12:11 | 15:18 | Major | Cloudflare Global Network experiencing issues |
| Udemy | 12:11 | 15:13 | Critical | Cloudfare incident - Site degraded |
| Paddle | 12:17 | 14:51 | Minor | Partial Checkout Issues |
| Alpaca | 12:50 | 20:04 | Critical | Alpaca Applications Down |
| Brex | 13:17 | 16:30 | Critical | Cloudflare Outage |
| Datadog | 13:17 | 13:53 | Minor | Delayed Monitors Notifications |
| Kinsta | 13:18 | 18:42 | Major | MyKinsta Dashboard Outage | Websites affected too |
| Atlassian Confluence | 13:32 | 16:57 | Critical | Elevated Errors in Confluence Whiteboards and Databases Due to Cloudflare Outage |
| Klaviyo | 13:58 | 19:06 | Major | Cloudflare incident disrupting onsite events |
| Figma | 14:01 | 18:04 | None set | Service Disruption with Sites and Make |
| Miro | 14:11 | 17:30 | Minor | Ongoing incident with one of our providers - Cloudflare |
| Dropbox | 14:38 | 14:53 | Major | DocSend impacted by ongoing Cloudflare outage |
Half the vendors were still red when Cloudflare was fine
Cloudflare says core traffic was largely back by 14:30 and all of its systems were normal at 17:06. The vendors' own pages lagged well behind. Of the 508 companies with a usable close time, 92 had closed their incident by 15:00. By 17:00 the count was 256, almost exactly half. 45 left theirs open into November 19 or later.
Status data can't tell you why each one stayed open, and some of it is just a page nobody came back to update. The usual reasons after an edge outage are real, though: queued jobs and webhooks that have to drain, caches that refill, retries that run into rate limits, and teams that sensibly wait a while before saying it's fixed. For a customer the effect is the same. "Cloudflare is back" did not mean your vendors were.
| Before 14:00 | 26 |
| 14:00 to 14:59 | 66 |
| 15:00 to 15:59 | 121 |
| 16:00 to 16:59 | 43 |
| 17:00 to 17:59 | 53 |
| 18:00 to 18:59 | 70 |
| 19:00 to 19:59 | 32 |
| 20:00 to 20:59 | 30 |
| 21:00 to 23:59 | 22 |
| November 19 or later | 45 |
December 5: the same pattern in 25 minutes
Seventeen days later Cloudflare had a second outage: Cloudflare outage on December 5, 2025. Cloudflare was raising the size of request bodies its firewall inspects, to protect customers from a React Server Components vulnerability (CVE-2025-55182). A related configuration change hit an old bug in its FL1 proxy. About 28% of Cloudflare's HTTP traffic got errors from 08:47 to 09:12 UTC.
185 companies named Cloudflare in an incident that day. They were much quicker this time, with a median of 22 minutes from the change to the vendor's post, which suggests a lot of teams still had their November playbook open. Even so, 65 of the 185 opened their incident after 09:12, when Cloudflare had already rolled the change back. Their customers heard about the outage after it was over.
96 companies named Cloudflare in both incidents. If your product depends on any of them, you were hit twice in 17 days by a provider you may never have signed a contract with.
What to change before the next one
Find out which of your vendors sit behind Cloudflare. Your contract list won't show it. Go through the vendors in your login, checkout, support and deploy paths and look at their response headers. A "cf-ray" header means the request went through Cloudflare. On November 18, 524 vendors in our catalog had that exposure, and most of their customers found out on the day.
Watch the upstream as well as the vendor. A vendor's status page lagged the real failure by a median of 89 minutes in November. When Cloudflare itself goes red, assume every Cloudflare-fronted vendor in your stack is affected until your own checks say otherwise, instead of waiting for each of them to confirm it.
Don't close your incident when Cloudflare closes theirs. Half the vendors were still red at 17:00. Keep yours open until the vendors your product needs have recovered, or until your own end-to-end checks pass.
Write the customer message in advance. Most of the 89 minutes went into finding the cause and agreeing on wording. A prepared note that says which features rely on outside providers, and where customers can check, cuts that down. The third-party outage runbook walks through who sends what, and when.
PulsAPI follows Cloudflare's status page and the status pages of the vendors you depend on, and alerts you when any of them changes. That won't make vendors post faster. It does put the upstream and every affected vendor in one place, so the first sign of a Cloudflare problem doesn't have to be a support ticket.
How we counted
The data comes from PulsAPI's records of public vendor status pages, pulled read-only on November 16, 2026. A company counts if it opened at least one incident on November 18, 2025 (UTC) whose title or description mentions Cloudflare. Cloudflare's own catalog entries (Cloudflare, Cloudflare Workers and R2) are excluded. For each company we use its first such incident of the day.
524 is a floor, not a full count. A company only appears if its status page is in our catalog today and still shows its history from November 2025, and plenty of vendors wrote "an upstream provider" or "our CDN" without naming Cloudflare. A mention also isn't proof of an outage, as Amplitude's post shows.
Closing figures use records whose close time is after their start and within 30 days; median durations also require a close within three days. The baseline covers October 1, 2025 to January 31, 2026, minus both outage days and the day after each: Cloudflare came up on 42 of those 119 days, for an average of 1.7 companies on the days it did. On December 5, four incidents that mention Cloudflare started before Cloudflare's change at 08:47. They are in the 185 but not in the timing figures, since they may be unrelated.
FAQ: the November 18, 2025 Cloudflare outage
What caused the Cloudflare outage on November 18, 2025? A database permissions change made a Bot Management feature file roughly double in size. The file went over a 200-feature limit in Cloudflare's proxy, which crashed and returned 5xx errors to visitors of sites behind Cloudflare. Cloudflare says it was not a cyberattack.
How long did the Cloudflare outage last? Core traffic failed from 11:20 UTC until about 14:30 UTC, a little over three hours. Cloudflare says all of its systems were normal again at 17:06 UTC, nearly six hours after the start. Many affected vendors kept their own incidents open longer than that.
Which services were affected by the Cloudflare outage? At least 524 companies named Cloudflare on their status pages that day, including Canva, Udemy, Pipedrive, Brex, Klaviyo, Atlassian Confluence, Dropbox's DocSend and Groq. Most were SaaS, fintech and marketing products rather than infrastructure providers.
Has Cloudflare had a major outage since? Yes. On December 5, 2025 a configuration change broke about 28% of Cloudflare's HTTP traffic for 25 minutes, and 185 companies named it on their status pages. Later incidents are in the Cloudflare outage history.
About the Author
Builds and operates PulsAPI: the per-vendor fetchers, the status pipeline, and the uptime maths behind the public pages.
Start monitoring your stack
Aggregate real-time operational data from every service your stack depends on into a single dashboard. Free for your three most critical vendors; unlimited from $29/mo.